Elvis Saravia Clones His Voice in 15 Seconds Using Cartesia Sonic 3.6

Summary

AI trainer Elvis Saravia shared a brief X post demonstrating that Cartesia's Sonic 3.6 voice model can clone his voice from a roughly 15-second recording in seconds and produce fluent Japanese. Saravia does not speak this language. The post consists of a single enthusiastic endorsement accompanied by a link to Cartesia's playground.

Why it matters

Cloning from a 15-second audio clip in under a second and generating output in 44 languages lowers barriers to multilingual content production. This also raises expectations regarding consent and deepfake risks. If the quality claims are verified, it could reshape localization processes for content creators without the need to hire native speakers. Furthermore, it indicates that voice synthesis latency is approaching real-time, a significant milestone for interactive applications.

Highlights

• According to Saravia's claim, a usable clone is obtained in just a few seconds from a 15-second source recording
• Sonic 3.6 supports 44 languages; Saravia demoed Japanese output, a language he does not speak natively
• The post contains no benchmarks, A/B comparisons, or independent verification

Critical take

This 'article' is actually a single promotional tweet that lacks technical details, side-by-side comparisons, and mentions of consent, watermarking, or misuse prevention measures. Therefore, it is concerning that it received a 54-point AI score and completely lacks a critical framework. Until third-party evaluations arrive, the term 'impressive' remains a subjective impression from a single influencer rather than a verified capability claim.